To understand the differences among reference curves for bone mineral density (BMD) for Chinese, Japanese, and American Caucasian women, we measured the BMD at the anteroposterior (AP) lumbar spine (L1-L4), lateral lumbar spine (L2-L4), hip (including the femoral neck, trochanter, intertrochanter, Ward's triangle, and total hip), and ultradistal forearm by the dual-energy X-ray absorptiometry (DXA) in a total of 2728 healthy Chinese women, aged 5-96 years. Documented BMD data for Japanese women and device manufacturer's BMD new reference databases (including the NHANES III dataset) for American Caucasian women were also used in this study. The cubic regression model was found to fit best in analyzing the age-associated variations of BMD at various sites in Chinese women, i.e., the equations had the largest coefficient of determination (R2). At the AP/Lat spine, trochanter, intertrochanter, and Ward's triangle, BMD reference curves for Chinese women were lower than those for Japanese or Caucasian women, while at the femoral neck, total hip, and ultradistal forearm, the reference curves for Chinese women were higher than those for Japanese women, with overlaps and crossing of the curves for some age spans in comparing the Chinese and Caucasian women. There were significant differences in the peak BMD (PBMD) at various sites among the Chinese, Japanese, and Caucasian women (P = 0.000). The PBMDs for Chinese women at the lumbar spine and various sites of the hip were 5.7% +/- 2.1% (mean +/- SD, range, 2.7-7.9%) lower than those for Japanese women and 5.1% +/- 2.7% (range, 0.5-7.2%) lower than those for Caucasian women; however, the PBMDs for Chinese women were 26.2% higher than those for Japanese women and 10% higher than those for Caucasian women at the ultradistal forearm. After the PBMD, average T-scores of Chinese women for losses at the AP lumbar spine with increasing age were nearly identical to those for Japanese women, but both were greater than those for Caucasian women. The average T-scores for BMD loss at various sites in Chinese women were higher than those for both Japanese and Caucasian women except at the femoral neck, where the T-scores of Chinese women were exceeded by those of both Japanese and Caucasian women. Estimated from the T-score curve of BMD loss, the age of osteoporosis occurrence at the femoral neck in Chinese women was about 10 years later than that in Japanese or Caucasian women; at the AP spine, Chinese women were similar to Japanese women; at the other sites, the age for occurrence of osteoporosis in Chinese women was about 5-15 years earlier than that in either Japanese or Caucasian women. There are differences in prevalence or odds ratio (OR) of osteoporosis at the same skeletal region for Chinese, Japanese, and Caucasian women aged > or = 50 years or at different skeletal regions in women of the same race. The prevalences of osteoporosis at various regions of the hip in Chinese women are 10.1-19.8% and ORs are 22.0-32.3, of which prevalence at the femoral neck is the lowest (10.1%); the prevalences of osteoporosis in Japanese women are 11.6-16.8% and ORs are 21.1-26.3, of which prevalence at the femoral neck is the lowest (11.6%); and the prevalences of osteoporosis in Caucasian women are 13.0-20.0% and ORs are 19.4-48.9, of which prevalence at the femoral neck is the highest (20%). In conclusion, racial differences in BMD reference curves, prevalences, and risks of osteoporosis at various skeletal sites exist among native Chinese, Japanese, and American Caucasian women.